>No, I think "I don't know" is the most honest and scientifically robust answer.
It will impress at least with its solemn humility, but in this case, you *do*
know something, and that information makes the most likely answer that there is
indeed something in that blob, and that something is probably chloride. This
general scientific perspective should be understandable. It is more honest to
model what you think is most likely rather than putting nothing, which has zero
likelihood. Nothing is something also; hence, dishonesty. I know that you have
written some software to do this type of thing automatically, and I guess you
must have thought about the cautious vs. highest-likelihood perspectives quite
a bit in so doing. I would guess then your software is very agnostic? Maybe you
could try to see what the effect of different confidence thresholds (like
requiring anomalous signal to identify halides(!)) would be on general model
quality, say along the lines of pdb-redo. I predict that generally model
quality will decrease with the high confidence cutoffs you recommend, and I
hereby lay down the gauntlet for you to undertake this arduous and
time-consuming project.
>For those who insist on annotating every density blob, UNX atoms are the PDB's
>officially supported method for doing so (unless this has changed recently),
>or UNK/UNL for unknown amino acids and ligands. These are not without their
>own problems but they at least make both the presence of an atom and the
>uncertainty about its identity explicit.
Okay, so why don’t we recommend that this chloride be modeled as UNX? What’s
the problem? (And why not make every water into UNX for good measure as well?)
JPK